ارزیابی مناسب DSGE های ​​اقتصاد باز کوچک

We describe a simple extension of the Monacelli (2005) small open economy model that incorporates a non-tradable good, habit persistence and price indexation. The empirical fit of eight different specifications of this model is then tested in a Bayesian framework using data for three small open economies; Australia, Canada, and New Zealand. The results show that the model with a non-tradable good fits the data better than the one-good model across all specifications considered. In contrast to Rabanal and Rubio-Ramarez (2005), we find that adding price indexation to either the one- or two-good model deteriorates overall empirical fit.

Much recent work in macroeconomics has been on the development of dynamic stochastic general equilibrium (DSGE) models for the analysis of monetary policy. Indeed, there is mounting evidence that these models are capable of matching business cycle dynamics as well as purely statistical models, such as VARs Smets and Smets (2004). Accordingly, central banks have begun to move towards models with strong microeconomic foundations, away from the older-generation models developed in the 1990s.1 Central bank modellers are thus being confronted with a variety of questions relating to the design of DSGE models, such as how to model the real exchange rate and whether or not to include habit formation and price indexation. This paper takes a step back from the more complicated DSGEs currently being developed by academics and central banks, and aims to find some normative results regarding the empirical fit of small open economy DSGEs. Specifically, we aim to assess whether adding a non-tradable good, habit formation, and price indexation to a small open economy DSGE improves its overall empirical fit. Differences between the behavior of tradable and non-tradable prices have been shown to be important in determining real exchange rate dynamics (De Gregorio et al., 1994, Engel, 1999 and Burnstein et al., 2005). There is also empirical evidence suggesting that modelling tradable and non-tradable prices separately can improve the forecasting performance of the Phillips curve (Matheson, 2006). Hence, the non-tradable sector is a key feature of some recent open economy DSGEs, such as those of Laxton and Pesenti, 2003 and Devereux et al., 2006. Habit formation and price indexation have also been found to be important in fitting DSGEs (Smets and Smets, 2004 and Christiano et al., 2005), and these mechanisms have become standard features of the models being developed by many central banks (see, for example, Murchison and Rennison (2006)). Since Gali and Gertler (1999) developed their popular New Keynesian DSGE for the analysis of monetary policy in a closed economy, the model has been extended to the small open economy by Gali and Monacelli (2005), and augmented further by allowing for deviations from the law of one price by Monacelli (2005). Acknowledging the importance of the non-tradable sector, Santacreu (2005) extended the Gali and Monacelli (2005) by adding a non-tradable good (as well as habit persistence and price indexation). Likewise, the Monacelli (2005) model has been augmented with habit persistence and price indexation by Justiniano and Preston (forthcoming) and Liu (2006). But the question remains: Which of these additional features are key to improving the overall empirical fit of the canonical small open economy DSGE?2 We outline a general model that allows for deviations from the law of one price, habit persistence, price indexation and a non-tradable good. This general model nests a variety of different specifications of the small open economy DSGE, such as Monacelli’s one-good model and versions of the one- and two-good model that exclude habit persistence and/or price indexation. Altogether, we have eight different specifications of the model. The empirical fit of the model is then tested in a Bayesian framework using data for three inflation targeting small open economies; Australia, Canada, and New Zealand.3 The results show that the two-good model fits the data better than the one-good model across all specifications considered. In contrast to the Rabanal and Rubio-Ramarez (2005) results for the closed economy, we also find that the addition of price indexation to either the one- or two-good model deteriorates overall empirical fit. Indeed, our results suggest that, if one were to augment the one- or two-good model with endogenous persistence mechanisms, better fit can be achieved by using habit formation rather than price indexation. The paper proceeds as follows. We first outline the general model specification in Section 2. Next, we discuss the Bayesian estimation methodology and the data. The results of the model comparison are discussed in Section 6, and we conclude in Section 7.

We outlined an extension of the Monacelli (2005) model that includes a non-tradable sector, habit persistence, and price indexation. The empirical fit of eight different specifications of this model were then tested in a Bayesian framework using data for three inflation targeting small open economies; Australia, Canada, and New Zealand. We found that a key difference between the one- and two-good models is in the estimated behavior of domestically-produced the home tradable good. Specifically, the estimated one-good model suggested a much larger home-produced tradable sector than the two-good model, and that the Calvo parameter estimates for inflation in the home-produced tradable sector differed notably across the one- and two-good models. In comparing the marginal likelihoods of the models, we found that the two-good model fits the data better than the one-good model across all specifications considered. It is clear that the existence of non-tradable goods violates the assumptions of the canonical small open economy model, thus deteriorating its ability to capture features of the data relative to a model that allows for non-tradable goods. Also, in contrast to the findings of Rabanal and Rubio-Ramarez (2005), we found that the addition of price indexation to either the one- or two-good model deteriorates overall empirical fit.